Entropy 2012, 14, 177-212; doi:10.3390/e14020177 OPEN ACCESS entropy ISSN 1099-4300 www.mdpi.com/journal/entropy Article Disentangling Complexity from Randomness and Chaos Lena C. Zuchowski Department of History and Philosophy of Science, University of Cambridge, Free School Lane, CB2 3RH Cambridge, UK; E-Mail: [email protected] Received: 27 December 2011; in revised form: 26 January 2012 / Accepted: 31 January 2012 / Published: 7 February 2012 Abstract: This study aims to disentangle complexity from randomness and chaos, and to present a definition of complexity that emphasizes its epistemically distinct qualities. I will review existing attempts at defining complexity and argue that these suffer from two major faults: a tendency to neglect the underlying dynamics and to focus exclusively on the phenomenology of complex systems; and linguistic imprecisions in describing these phenomenologies. I will argue that the tendency to discuss phenomenology removed from the underlying dynamics is the main root of the difficulties in distinguishing complex from chaotic or random systems. In my own definition, I will explicitly try to avoid these pitfalls. The theoretical contemplations in this paper will be tested on a sample of five models: the random Kac ring, the chaotic CA30, the regular CA90, the complex CA110 and the complex Bak-Sneppen model. Although these modelling studies are restricted in scope and can only be seen as preliminary, they still constitute on of the first attempts to investigate complex systems comparatively. Keywords: complexity; complexity definition; entropy 1. Introduction “Was sich uberhaupt¨ sagen laßt,¨ laßt¨ sich klar sagen; und wovon man nicht reden kann, daruber¨ muß man schweigen.” (L. Wittgenstein, Tractatus Logico-Philosophicus) During the last ten years complexity research has received a relatively large amount of attention by both the scientific community and the general public: Paramount scientific figures like Nobel Laureate Entropy 2012, 14 178 Murray Gell-Mann have championed its cause (e.g., [1]); Science magazine devoted a special issue to it; and in the 2004’s bestseller lists, The Swarm [2] explored the sinister consequences of not paying enough attention to complex systems. One of the greatest draws of complexity as a field of research is the possibility of recognizing it in virtually every branch of science and the social sciences (e.g., [3]). However, despite the labelling of an increasingly large number of models and natural systems as “complex”, the definition of the term has remained vague. Standish [4] pointed out that these difficulties extend to both a qualitative identification of a conclusive set of defining features that a complex system should possess, as well as to the quantitative measurements of complexity as a property. The difficulty of finding an unequivocal conceptualization and measure of complexity is also recognized by Gell-Mann [5], who states that (p. 1): “[A] variety of different measures would be required to capture all our intuitive ideas about what is meant by complexity and by its opposite, simplicity.” In fact, Lloyd [6] enumerates forty-two different existing complexity measures in what is described as “a nonexhaustive list” (p. 7). Despite the lack of agreement on the definition of complexity, complexity scientists have been eager to stress the importance of the concept as central to nature. For example, Wolfram [7] writes about his work: “Perhaps the most dramatic [benefit] is that it yields a resolution to what has long been considered the single greatest mystery of the natural world: what secret is it that allows nature seemingly so effortlessly to produce so much that appears to us complex.” (p. 2) However, the failure to precisely state what it actually is that needs to be studied in order to unravel “nature’s greatest mystery” has drawn extensive criticism. Horgan [8] describes complexity as an unjustifiably hyped “pop-science movement” that spun off from chaos theory. A closer review of the complexity measures compiled by Lloyd [6] reveals that they indeed borrow heavily from a number of “ancestor theories” like statistical mechanics, chaos theory and computational physics. Likewise, a closer look at the conception of these measures reveals that the struggle for a definition of complexity is fueled by two factors: a failure to fully emancipate oneself from these epistemological ancestors and the uncritical use of metaphorical and imprecise language. This paper aims to present a structured account of the “intuitive ideas of what is meant by complexity” mentioned by Gell-Mann [5]. With the possible exception of the currently mostly speculative work by Wolfram [7], intuitions about complexity have been largely restricted to its phenomenology. I will argue that the tendency to discuss phenomenology removed from the underlying dynamics is the main root of the difficulties in distinguishing chaotic from complex systems. I will claim that a purely dynamical definition of complexity is possible, however, it would be more inclusive than the complexity community wishes it to be. Instead, a phenomenological sieve must be imposed to distinguish the “interesting” systems from “simple” ones with similar dynamics. Secondly, I will show that both previous attempts to design such a sieve as well as the general complexity discourse suffer from either semantic circularities or an over-reliance on metaphorical language, which seems to compensate for a potentially fundamental difficulty in arriving at an objective Entropy 2012, 14 179 description. It will also be shown that quantitative complexity measures based on the most prevalent metaphors are not successful in identifying complex models as such. As indicated by the mildly ironic motto of this paper I believe that it would have aided the the progress of complexity science if the community had refrained from hiding the inability to recognize the essence of complexity behind eye-catching but fundamentally imprecise phrases. Based on the insights gained in the first parts of this paper, we will finally design our own complexity definition. Methodologically, I complement my theoretical investigation by the analysis of a sample of five simple models. The phenomenologies of two of these are generally judged to be complex: the Wolfram Cellular Automaton (CA) 110 (e.g., [7]) and a discrete Bak-Sneppen model [9]. I have deliberately sought to include models that are based on very different dynamical premises (e.g., the CA110 is deterministic while the Bak-Sneppen model is probabilistic), thereby hoping to address a lacuna in comparative complexity studies. In addition to these I will consider: the Wolfram CA30, a model with pseudo-random (often called “chaotic”) output; the Kac ring, a dynamically random and phenomenologically entropy-maximizing model; and the Wolfram CA90, a model with regular phenomenology. The models will be used as both a means of verifying existing assumptions about complexity as well as a testbed for my own ideas. Underlying the reliance on these simple computer models is the assumption that complexity science can still very much be identified with the study of such simulations (e.g., [10]). In Section 2 I will describe the models used in this study. Section 3 contains a discussion of dynamical complexity definitions and measures. In Section 4 I will examine qualitative and quantitative descriptions of complex phenomenology. Conclusions will be drawn in Section 5, which also contains my attempt at a rudimentary complexity definition. 2. Five Simple Models In this section I will describe the five simple models used in this study. Wherever possible, the models have been chosen as typical, or even well-known, representatives of a larger class of such simulations. I have also deliberately included both a deterministic as well as probabilistic complex model. 2.1. The Kac Ring The Kac model was developed by Kac [11] as a means of demonstrating that coarse graining and Stosszahlansatz-like dynamical assumptions can lead to irreversible evolution of a macroscopic entropy. It describes an idealized system of particles whose phase-space values are determined by randomized collisions only. Dorfmann [12] and Gottwald and Oliver [13] provide a detailed discussion of the dynamics of the system in the context of thermodynamics and statistical physics. My abbreviated account below is based on their exposition. The KAC model consists of a one-dimensional, periodic lattice of N sites arranged around a ring. Each site is occupied by a particle, which has either the property “black” or the property “white”. Neighbouring sites are joined by edges. Each of the N edges is assigned one of two markers, which will be denoted as S and S¯. Entropy 2012, 14 180 The dynamics of the system is discrete and consists of particles moving clockwise to the neighbouring site. Thereby, the edge markers control the evolution of the colour property: a particle changes colour if it traverses an edge marked S¯ but remains unchanged if the edge marker is S. If the same number of steps is retraced in the counter-clockwise direction, the initial state is reached again. The dynamics of the ring is therefore reversible. Furthermore, the time evolution of the system is periodic. Depending on the number n of colour-changing edges S¯, states reoccur with a period of N or 2N time-steps for even or odd n, respectively. Mimicking the dual mode of description employed in statistical physics and thermodynamics, microscopic and macroscopic variables will be used to describe the Kac model. On the macroscopic level, the number B of black particles and the number W of white particles will be used as fundamental variables. On a mesoscale level, convenient properties to consider are the number of black or white particles before an S¯ edge: b and w, respectively. From the previous paragraph, the importance of these quantities for the evolution of the model in the next step is immediately apparent: b is equal to the number of particles changing from black to white and w is equal to the number of particles changing from white to black.